---
license: apache-2.0
datasets:
- ustc-zyt/time-r1-data
language:
- en
metrics:
- mse
- mae
base_model:
- Qwen/Qwen2.5-7B
---
# 🧠 Time-R1 Reinforced Model Weights

These are the official **reinforcement learning (RL) fine-tuned model checkpoints** for the paper:
**"Time Series Forecasting as Reasoning: A Slow-Thinking Approach with Reinforced LLMs"**.

---

## 📦 Model Details

* **Base Model**: Qwen2.5-7B
* **Tuning Framework**: [Verl](https://github.com/volcengine/verl) + [LLaMA Factory](https://github.com/hiyouga/LLaMA-Factory)
* **Final Stage**: Trained using GRIP (Group-based Relative Importance Policy optimization)
* **Objective**: Multi-horizon time series forecasting with structured reasoning

---

## 📦 Files Included

This model follows the standard Hugging Face `transformers` format and uses the efficient `safetensors` backend.

```
Time-R1/
├── config.json
├── generation_config.json
├── model.safetensors.index.json
├── model-00001-of-00004.safetensors
├── model-00002-of-00004.safetensors
├── model-00003-of-00004.safetensors
├── model-00004-of-00004.safetensors
├── tokenizer_config.json
├── tokenizer.json
└── vocab.json
```

✅ Fully compatible with Hugging Face `transformers` and `AutoModelForCausalLM`.